xref: /openbmc/linux/net/bpf/test_run.c (revision b0bc615df488abd0e95107e4a9ecefb9bf8c250a)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Copyright (c) 2017 Facebook
3  */
4 #include <linux/bpf.h>
5 #include <linux/btf.h>
6 #include <linux/btf_ids.h>
7 #include <linux/slab.h>
8 #include <linux/init.h>
9 #include <linux/vmalloc.h>
10 #include <linux/etherdevice.h>
11 #include <linux/filter.h>
12 #include <linux/rcupdate_trace.h>
13 #include <linux/sched/signal.h>
14 #include <net/bpf_sk_storage.h>
15 #include <net/sock.h>
16 #include <net/tcp.h>
17 #include <net/net_namespace.h>
18 #include <net/page_pool.h>
19 #include <linux/error-injection.h>
20 #include <linux/smp.h>
21 #include <linux/sock_diag.h>
22 #include <net/xdp.h>
23 
24 #define CREATE_TRACE_POINTS
25 #include <trace/events/bpf_test_run.h>
26 
27 struct bpf_test_timer {
28 	enum { NO_PREEMPT, NO_MIGRATE } mode;
29 	u32 i;
30 	u64 time_start, time_spent;
31 };
32 
33 static void bpf_test_timer_enter(struct bpf_test_timer *t)
34 	__acquires(rcu)
35 {
36 	rcu_read_lock();
37 	if (t->mode == NO_PREEMPT)
38 		preempt_disable();
39 	else
40 		migrate_disable();
41 
42 	t->time_start = ktime_get_ns();
43 }
44 
45 static void bpf_test_timer_leave(struct bpf_test_timer *t)
46 	__releases(rcu)
47 {
48 	t->time_start = 0;
49 
50 	if (t->mode == NO_PREEMPT)
51 		preempt_enable();
52 	else
53 		migrate_enable();
54 	rcu_read_unlock();
55 }
56 
57 static bool bpf_test_timer_continue(struct bpf_test_timer *t, int iterations,
58 				    u32 repeat, int *err, u32 *duration)
59 	__must_hold(rcu)
60 {
61 	t->i += iterations;
62 	if (t->i >= repeat) {
63 		/* We're done. */
64 		t->time_spent += ktime_get_ns() - t->time_start;
65 		do_div(t->time_spent, t->i);
66 		*duration = t->time_spent > U32_MAX ? U32_MAX : (u32)t->time_spent;
67 		*err = 0;
68 		goto reset;
69 	}
70 
71 	if (signal_pending(current)) {
72 		/* During iteration: we've been cancelled, abort. */
73 		*err = -EINTR;
74 		goto reset;
75 	}
76 
77 	if (need_resched()) {
78 		/* During iteration: we need to reschedule between runs. */
79 		t->time_spent += ktime_get_ns() - t->time_start;
80 		bpf_test_timer_leave(t);
81 		cond_resched();
82 		bpf_test_timer_enter(t);
83 	}
84 
85 	/* Do another round. */
86 	return true;
87 
88 reset:
89 	t->i = 0;
90 	return false;
91 }
92 
93 /* We put this struct at the head of each page with a context and frame
94  * initialised when the page is allocated, so we don't have to do this on each
95  * repetition of the test run.
96  */
97 struct xdp_page_head {
98 	struct xdp_buff orig_ctx;
99 	struct xdp_buff ctx;
100 	union {
101 		/* ::data_hard_start starts here */
102 		DECLARE_FLEX_ARRAY(struct xdp_frame, frame);
103 		DECLARE_FLEX_ARRAY(u8, data);
104 	};
105 };
106 
107 struct xdp_test_data {
108 	struct xdp_buff *orig_ctx;
109 	struct xdp_rxq_info rxq;
110 	struct net_device *dev;
111 	struct page_pool *pp;
112 	struct xdp_frame **frames;
113 	struct sk_buff **skbs;
114 	struct xdp_mem_info mem;
115 	u32 batch_size;
116 	u32 frame_cnt;
117 };
118 
119 /* tools/testing/selftests/bpf/prog_tests/xdp_do_redirect.c:%MAX_PKT_SIZE
120  * must be updated accordingly this gets changed, otherwise BPF selftests
121  * will fail.
122  */
123 #define TEST_XDP_FRAME_SIZE (PAGE_SIZE - sizeof(struct xdp_page_head))
124 #define TEST_XDP_MAX_BATCH 256
125 
126 static void xdp_test_run_init_page(struct page *page, void *arg)
127 {
128 	struct xdp_page_head *head = phys_to_virt(page_to_phys(page));
129 	struct xdp_buff *new_ctx, *orig_ctx;
130 	u32 headroom = XDP_PACKET_HEADROOM;
131 	struct xdp_test_data *xdp = arg;
132 	size_t frm_len, meta_len;
133 	struct xdp_frame *frm;
134 	void *data;
135 
136 	orig_ctx = xdp->orig_ctx;
137 	frm_len = orig_ctx->data_end - orig_ctx->data_meta;
138 	meta_len = orig_ctx->data - orig_ctx->data_meta;
139 	headroom -= meta_len;
140 
141 	new_ctx = &head->ctx;
142 	frm = head->frame;
143 	data = head->data;
144 	memcpy(data + headroom, orig_ctx->data_meta, frm_len);
145 
146 	xdp_init_buff(new_ctx, TEST_XDP_FRAME_SIZE, &xdp->rxq);
147 	xdp_prepare_buff(new_ctx, data, headroom, frm_len, true);
148 	new_ctx->data = new_ctx->data_meta + meta_len;
149 
150 	xdp_update_frame_from_buff(new_ctx, frm);
151 	frm->mem = new_ctx->rxq->mem;
152 
153 	memcpy(&head->orig_ctx, new_ctx, sizeof(head->orig_ctx));
154 }
155 
156 static int xdp_test_run_setup(struct xdp_test_data *xdp, struct xdp_buff *orig_ctx)
157 {
158 	struct page_pool *pp;
159 	int err = -ENOMEM;
160 	struct page_pool_params pp_params = {
161 		.order = 0,
162 		.flags = 0,
163 		.pool_size = xdp->batch_size,
164 		.nid = NUMA_NO_NODE,
165 		.init_callback = xdp_test_run_init_page,
166 		.init_arg = xdp,
167 	};
168 
169 	xdp->frames = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
170 	if (!xdp->frames)
171 		return -ENOMEM;
172 
173 	xdp->skbs = kvmalloc_array(xdp->batch_size, sizeof(void *), GFP_KERNEL);
174 	if (!xdp->skbs)
175 		goto err_skbs;
176 
177 	pp = page_pool_create(&pp_params);
178 	if (IS_ERR(pp)) {
179 		err = PTR_ERR(pp);
180 		goto err_pp;
181 	}
182 
183 	/* will copy 'mem.id' into pp->xdp_mem_id */
184 	err = xdp_reg_mem_model(&xdp->mem, MEM_TYPE_PAGE_POOL, pp);
185 	if (err)
186 		goto err_mmodel;
187 
188 	xdp->pp = pp;
189 
190 	/* We create a 'fake' RXQ referencing the original dev, but with an
191 	 * xdp_mem_info pointing to our page_pool
192 	 */
193 	xdp_rxq_info_reg(&xdp->rxq, orig_ctx->rxq->dev, 0, 0);
194 	xdp->rxq.mem.type = MEM_TYPE_PAGE_POOL;
195 	xdp->rxq.mem.id = pp->xdp_mem_id;
196 	xdp->dev = orig_ctx->rxq->dev;
197 	xdp->orig_ctx = orig_ctx;
198 
199 	return 0;
200 
201 err_mmodel:
202 	page_pool_destroy(pp);
203 err_pp:
204 	kvfree(xdp->skbs);
205 err_skbs:
206 	kvfree(xdp->frames);
207 	return err;
208 }
209 
210 static void xdp_test_run_teardown(struct xdp_test_data *xdp)
211 {
212 	xdp_unreg_mem_model(&xdp->mem);
213 	page_pool_destroy(xdp->pp);
214 	kfree(xdp->frames);
215 	kfree(xdp->skbs);
216 }
217 
218 static bool frame_was_changed(const struct xdp_page_head *head)
219 {
220 	/* xdp_scrub_frame() zeroes the data pointer, flags is the last field,
221 	 * i.e. has the highest chances to be overwritten. If those two are
222 	 * untouched, it's most likely safe to skip the context reset.
223 	 */
224 	return head->frame->data != head->orig_ctx.data ||
225 	       head->frame->flags != head->orig_ctx.flags;
226 }
227 
228 static bool ctx_was_changed(struct xdp_page_head *head)
229 {
230 	return head->orig_ctx.data != head->ctx.data ||
231 		head->orig_ctx.data_meta != head->ctx.data_meta ||
232 		head->orig_ctx.data_end != head->ctx.data_end;
233 }
234 
235 static void reset_ctx(struct xdp_page_head *head)
236 {
237 	if (likely(!frame_was_changed(head) && !ctx_was_changed(head)))
238 		return;
239 
240 	head->ctx.data = head->orig_ctx.data;
241 	head->ctx.data_meta = head->orig_ctx.data_meta;
242 	head->ctx.data_end = head->orig_ctx.data_end;
243 	xdp_update_frame_from_buff(&head->ctx, head->frame);
244 }
245 
246 static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
247 			   struct sk_buff **skbs,
248 			   struct net_device *dev)
249 {
250 	gfp_t gfp = __GFP_ZERO | GFP_ATOMIC;
251 	int i, n;
252 	LIST_HEAD(list);
253 
254 	n = kmem_cache_alloc_bulk(skbuff_cache, gfp, nframes, (void **)skbs);
255 	if (unlikely(n == 0)) {
256 		for (i = 0; i < nframes; i++)
257 			xdp_return_frame(frames[i]);
258 		return -ENOMEM;
259 	}
260 
261 	for (i = 0; i < nframes; i++) {
262 		struct xdp_frame *xdpf = frames[i];
263 		struct sk_buff *skb = skbs[i];
264 
265 		skb = __xdp_build_skb_from_frame(xdpf, skb, dev);
266 		if (!skb) {
267 			xdp_return_frame(xdpf);
268 			continue;
269 		}
270 
271 		list_add_tail(&skb->list, &list);
272 	}
273 	netif_receive_skb_list(&list);
274 
275 	return 0;
276 }
277 
278 static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
279 			      u32 repeat)
280 {
281 	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
282 	int err = 0, act, ret, i, nframes = 0, batch_sz;
283 	struct xdp_frame **frames = xdp->frames;
284 	struct xdp_page_head *head;
285 	struct xdp_frame *frm;
286 	bool redirect = false;
287 	struct xdp_buff *ctx;
288 	struct page *page;
289 
290 	batch_sz = min_t(u32, repeat, xdp->batch_size);
291 
292 	local_bh_disable();
293 	xdp_set_return_frame_no_direct();
294 
295 	for (i = 0; i < batch_sz; i++) {
296 		page = page_pool_dev_alloc_pages(xdp->pp);
297 		if (!page) {
298 			err = -ENOMEM;
299 			goto out;
300 		}
301 
302 		head = phys_to_virt(page_to_phys(page));
303 		reset_ctx(head);
304 		ctx = &head->ctx;
305 		frm = head->frame;
306 		xdp->frame_cnt++;
307 
308 		act = bpf_prog_run_xdp(prog, ctx);
309 
310 		/* if program changed pkt bounds we need to update the xdp_frame */
311 		if (unlikely(ctx_was_changed(head))) {
312 			ret = xdp_update_frame_from_buff(ctx, frm);
313 			if (ret) {
314 				xdp_return_buff(ctx);
315 				continue;
316 			}
317 		}
318 
319 		switch (act) {
320 		case XDP_TX:
321 			/* we can't do a real XDP_TX since we're not in the
322 			 * driver, so turn it into a REDIRECT back to the same
323 			 * index
324 			 */
325 			ri->tgt_index = xdp->dev->ifindex;
326 			ri->map_id = INT_MAX;
327 			ri->map_type = BPF_MAP_TYPE_UNSPEC;
328 			fallthrough;
329 		case XDP_REDIRECT:
330 			redirect = true;
331 			ret = xdp_do_redirect_frame(xdp->dev, ctx, frm, prog);
332 			if (ret)
333 				xdp_return_buff(ctx);
334 			break;
335 		case XDP_PASS:
336 			frames[nframes++] = frm;
337 			break;
338 		default:
339 			bpf_warn_invalid_xdp_action(NULL, prog, act);
340 			fallthrough;
341 		case XDP_DROP:
342 			xdp_return_buff(ctx);
343 			break;
344 		}
345 	}
346 
347 out:
348 	if (redirect)
349 		xdp_do_flush();
350 	if (nframes) {
351 		ret = xdp_recv_frames(frames, nframes, xdp->skbs, xdp->dev);
352 		if (ret)
353 			err = ret;
354 	}
355 
356 	xdp_clear_return_frame_no_direct();
357 	local_bh_enable();
358 	return err;
359 }
360 
361 static int bpf_test_run_xdp_live(struct bpf_prog *prog, struct xdp_buff *ctx,
362 				 u32 repeat, u32 batch_size, u32 *time)
363 
364 {
365 	struct xdp_test_data xdp = { .batch_size = batch_size };
366 	struct bpf_test_timer t = { .mode = NO_MIGRATE };
367 	int ret;
368 
369 	if (!repeat)
370 		repeat = 1;
371 
372 	ret = xdp_test_run_setup(&xdp, ctx);
373 	if (ret)
374 		return ret;
375 
376 	bpf_test_timer_enter(&t);
377 	do {
378 		xdp.frame_cnt = 0;
379 		ret = xdp_test_run_batch(&xdp, prog, repeat - t.i);
380 		if (unlikely(ret < 0))
381 			break;
382 	} while (bpf_test_timer_continue(&t, xdp.frame_cnt, repeat, &ret, time));
383 	bpf_test_timer_leave(&t);
384 
385 	xdp_test_run_teardown(&xdp);
386 	return ret;
387 }
388 
389 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
390 			u32 *retval, u32 *time, bool xdp)
391 {
392 	struct bpf_prog_array_item item = {.prog = prog};
393 	struct bpf_run_ctx *old_ctx;
394 	struct bpf_cg_run_ctx run_ctx;
395 	struct bpf_test_timer t = { NO_MIGRATE };
396 	enum bpf_cgroup_storage_type stype;
397 	int ret;
398 
399 	for_each_cgroup_storage_type(stype) {
400 		item.cgroup_storage[stype] = bpf_cgroup_storage_alloc(prog, stype);
401 		if (IS_ERR(item.cgroup_storage[stype])) {
402 			item.cgroup_storage[stype] = NULL;
403 			for_each_cgroup_storage_type(stype)
404 				bpf_cgroup_storage_free(item.cgroup_storage[stype]);
405 			return -ENOMEM;
406 		}
407 	}
408 
409 	if (!repeat)
410 		repeat = 1;
411 
412 	bpf_test_timer_enter(&t);
413 	old_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
414 	do {
415 		run_ctx.prog_item = &item;
416 		local_bh_disable();
417 		if (xdp)
418 			*retval = bpf_prog_run_xdp(prog, ctx);
419 		else
420 			*retval = bpf_prog_run(prog, ctx);
421 		local_bh_enable();
422 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, time));
423 	bpf_reset_run_ctx(old_ctx);
424 	bpf_test_timer_leave(&t);
425 
426 	for_each_cgroup_storage_type(stype)
427 		bpf_cgroup_storage_free(item.cgroup_storage[stype]);
428 
429 	return ret;
430 }
431 
432 static int bpf_test_finish(const union bpf_attr *kattr,
433 			   union bpf_attr __user *uattr, const void *data,
434 			   struct skb_shared_info *sinfo, u32 size,
435 			   u32 retval, u32 duration)
436 {
437 	void __user *data_out = u64_to_user_ptr(kattr->test.data_out);
438 	int err = -EFAULT;
439 	u32 copy_size = size;
440 
441 	/* Clamp copy if the user has provided a size hint, but copy the full
442 	 * buffer if not to retain old behaviour.
443 	 */
444 	if (kattr->test.data_size_out &&
445 	    copy_size > kattr->test.data_size_out) {
446 		copy_size = kattr->test.data_size_out;
447 		err = -ENOSPC;
448 	}
449 
450 	if (data_out) {
451 		int len = sinfo ? copy_size - sinfo->xdp_frags_size : copy_size;
452 
453 		if (len < 0) {
454 			err = -ENOSPC;
455 			goto out;
456 		}
457 
458 		if (copy_to_user(data_out, data, len))
459 			goto out;
460 
461 		if (sinfo) {
462 			int i, offset = len;
463 			u32 data_len;
464 
465 			for (i = 0; i < sinfo->nr_frags; i++) {
466 				skb_frag_t *frag = &sinfo->frags[i];
467 
468 				if (offset >= copy_size) {
469 					err = -ENOSPC;
470 					break;
471 				}
472 
473 				data_len = min_t(u32, copy_size - offset,
474 						 skb_frag_size(frag));
475 
476 				if (copy_to_user(data_out + offset,
477 						 skb_frag_address(frag),
478 						 data_len))
479 					goto out;
480 
481 				offset += data_len;
482 			}
483 		}
484 	}
485 
486 	if (copy_to_user(&uattr->test.data_size_out, &size, sizeof(size)))
487 		goto out;
488 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
489 		goto out;
490 	if (copy_to_user(&uattr->test.duration, &duration, sizeof(duration)))
491 		goto out;
492 	if (err != -ENOSPC)
493 		err = 0;
494 out:
495 	trace_bpf_test_finish(&err);
496 	return err;
497 }
498 
499 /* Integer types of various sizes and pointer combinations cover variety of
500  * architecture dependent calling conventions. 7+ can be supported in the
501  * future.
502  */
503 __diag_push();
504 __diag_ignore_all("-Wmissing-prototypes",
505 		  "Global functions as their definitions will be in vmlinux BTF");
506 __bpf_kfunc int bpf_fentry_test1(int a)
507 {
508 	return a + 1;
509 }
510 EXPORT_SYMBOL_GPL(bpf_fentry_test1);
511 
512 int noinline bpf_fentry_test2(int a, u64 b)
513 {
514 	return a + b;
515 }
516 
517 int noinline bpf_fentry_test3(char a, int b, u64 c)
518 {
519 	return a + b + c;
520 }
521 
522 int noinline bpf_fentry_test4(void *a, char b, int c, u64 d)
523 {
524 	return (long)a + b + c + d;
525 }
526 
527 int noinline bpf_fentry_test5(u64 a, void *b, short c, int d, u64 e)
528 {
529 	return a + (long)b + c + d + e;
530 }
531 
532 int noinline bpf_fentry_test6(u64 a, void *b, short c, int d, void *e, u64 f)
533 {
534 	return a + (long)b + c + d + (long)e + f;
535 }
536 
537 struct bpf_fentry_test_t {
538 	struct bpf_fentry_test_t *a;
539 };
540 
541 int noinline bpf_fentry_test7(struct bpf_fentry_test_t *arg)
542 {
543 	return (long)arg;
544 }
545 
546 int noinline bpf_fentry_test8(struct bpf_fentry_test_t *arg)
547 {
548 	return (long)arg->a;
549 }
550 
551 __bpf_kfunc u32 bpf_fentry_test9(u32 *a)
552 {
553 	return *a;
554 }
555 
556 __bpf_kfunc int bpf_modify_return_test(int a, int *b)
557 {
558 	*b += 1;
559 	return a + *b;
560 }
561 
562 __bpf_kfunc u64 bpf_kfunc_call_test1(struct sock *sk, u32 a, u64 b, u32 c, u64 d)
563 {
564 	return a + b + c + d;
565 }
566 
567 __bpf_kfunc int bpf_kfunc_call_test2(struct sock *sk, u32 a, u32 b)
568 {
569 	return a + b;
570 }
571 
572 __bpf_kfunc struct sock *bpf_kfunc_call_test3(struct sock *sk)
573 {
574 	return sk;
575 }
576 
577 long noinline bpf_kfunc_call_test4(signed char a, short b, int c, long d)
578 {
579 	/* Provoke the compiler to assume that the caller has sign-extended a,
580 	 * b and c on platforms where this is required (e.g. s390x).
581 	 */
582 	return (long)a + (long)b + (long)c + d;
583 }
584 
585 int noinline bpf_fentry_shadow_test(int a)
586 {
587 	return a + 1;
588 }
589 
590 struct prog_test_member1 {
591 	int a;
592 };
593 
594 struct prog_test_member {
595 	struct prog_test_member1 m;
596 	int c;
597 };
598 
599 struct prog_test_ref_kfunc {
600 	int a;
601 	int b;
602 	struct prog_test_member memb;
603 	struct prog_test_ref_kfunc *next;
604 	refcount_t cnt;
605 };
606 
607 static struct prog_test_ref_kfunc prog_test_struct = {
608 	.a = 42,
609 	.b = 108,
610 	.next = &prog_test_struct,
611 	.cnt = REFCOUNT_INIT(1),
612 };
613 
614 __bpf_kfunc struct prog_test_ref_kfunc *
615 bpf_kfunc_call_test_acquire(unsigned long *scalar_ptr)
616 {
617 	refcount_inc(&prog_test_struct.cnt);
618 	return &prog_test_struct;
619 }
620 
621 __bpf_kfunc void bpf_kfunc_call_test_offset(struct prog_test_ref_kfunc *p)
622 {
623 	WARN_ON_ONCE(1);
624 }
625 
626 __bpf_kfunc struct prog_test_member *
627 bpf_kfunc_call_memb_acquire(void)
628 {
629 	WARN_ON_ONCE(1);
630 	return NULL;
631 }
632 
633 __bpf_kfunc void bpf_kfunc_call_test_release(struct prog_test_ref_kfunc *p)
634 {
635 	refcount_dec(&p->cnt);
636 }
637 
638 __bpf_kfunc void bpf_kfunc_call_memb_release(struct prog_test_member *p)
639 {
640 }
641 
642 __bpf_kfunc void bpf_kfunc_call_memb1_release(struct prog_test_member1 *p)
643 {
644 	WARN_ON_ONCE(1);
645 }
646 
647 static int *__bpf_kfunc_call_test_get_mem(struct prog_test_ref_kfunc *p, const int size)
648 {
649 	if (size > 2 * sizeof(int))
650 		return NULL;
651 
652 	return (int *)p;
653 }
654 
655 __bpf_kfunc int *bpf_kfunc_call_test_get_rdwr_mem(struct prog_test_ref_kfunc *p,
656 						  const int rdwr_buf_size)
657 {
658 	return __bpf_kfunc_call_test_get_mem(p, rdwr_buf_size);
659 }
660 
661 __bpf_kfunc int *bpf_kfunc_call_test_get_rdonly_mem(struct prog_test_ref_kfunc *p,
662 						    const int rdonly_buf_size)
663 {
664 	return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size);
665 }
666 
667 /* the next 2 ones can't be really used for testing expect to ensure
668  * that the verifier rejects the call.
669  * Acquire functions must return struct pointers, so these ones are
670  * failing.
671  */
672 __bpf_kfunc int *bpf_kfunc_call_test_acq_rdonly_mem(struct prog_test_ref_kfunc *p,
673 						    const int rdonly_buf_size)
674 {
675 	return __bpf_kfunc_call_test_get_mem(p, rdonly_buf_size);
676 }
677 
678 __bpf_kfunc void bpf_kfunc_call_int_mem_release(int *p)
679 {
680 }
681 
682 __bpf_kfunc struct prog_test_ref_kfunc *
683 bpf_kfunc_call_test_kptr_get(struct prog_test_ref_kfunc **pp, int a, int b)
684 {
685 	struct prog_test_ref_kfunc *p = READ_ONCE(*pp);
686 
687 	if (!p)
688 		return NULL;
689 	refcount_inc(&p->cnt);
690 	return p;
691 }
692 
693 struct prog_test_pass1 {
694 	int x0;
695 	struct {
696 		int x1;
697 		struct {
698 			int x2;
699 			struct {
700 				int x3;
701 			};
702 		};
703 	};
704 };
705 
706 struct prog_test_pass2 {
707 	int len;
708 	short arr1[4];
709 	struct {
710 		char arr2[4];
711 		unsigned long arr3[8];
712 	} x;
713 };
714 
715 struct prog_test_fail1 {
716 	void *p;
717 	int x;
718 };
719 
720 struct prog_test_fail2 {
721 	int x8;
722 	struct prog_test_pass1 x;
723 };
724 
725 struct prog_test_fail3 {
726 	int len;
727 	char arr1[2];
728 	char arr2[];
729 };
730 
731 __bpf_kfunc void bpf_kfunc_call_test_pass_ctx(struct __sk_buff *skb)
732 {
733 }
734 
735 __bpf_kfunc void bpf_kfunc_call_test_pass1(struct prog_test_pass1 *p)
736 {
737 }
738 
739 __bpf_kfunc void bpf_kfunc_call_test_pass2(struct prog_test_pass2 *p)
740 {
741 }
742 
743 __bpf_kfunc void bpf_kfunc_call_test_fail1(struct prog_test_fail1 *p)
744 {
745 }
746 
747 __bpf_kfunc void bpf_kfunc_call_test_fail2(struct prog_test_fail2 *p)
748 {
749 }
750 
751 __bpf_kfunc void bpf_kfunc_call_test_fail3(struct prog_test_fail3 *p)
752 {
753 }
754 
755 __bpf_kfunc void bpf_kfunc_call_test_mem_len_pass1(void *mem, int mem__sz)
756 {
757 }
758 
759 __bpf_kfunc void bpf_kfunc_call_test_mem_len_fail1(void *mem, int len)
760 {
761 }
762 
763 __bpf_kfunc void bpf_kfunc_call_test_mem_len_fail2(u64 *mem, int len)
764 {
765 }
766 
767 __bpf_kfunc void bpf_kfunc_call_test_ref(struct prog_test_ref_kfunc *p)
768 {
769 	/* p != NULL, but p->cnt could be 0 */
770 }
771 
772 __bpf_kfunc void bpf_kfunc_call_test_destructive(void)
773 {
774 }
775 
776 __bpf_kfunc static u32 bpf_kfunc_call_test_static_unused_arg(u32 arg, u32 unused)
777 {
778 	return arg;
779 }
780 
781 __diag_pop();
782 
783 BTF_SET8_START(bpf_test_modify_return_ids)
784 BTF_ID_FLAGS(func, bpf_modify_return_test)
785 BTF_ID_FLAGS(func, bpf_fentry_test1, KF_SLEEPABLE)
786 BTF_SET8_END(bpf_test_modify_return_ids)
787 
788 static const struct btf_kfunc_id_set bpf_test_modify_return_set = {
789 	.owner = THIS_MODULE,
790 	.set   = &bpf_test_modify_return_ids,
791 };
792 
793 BTF_SET8_START(test_sk_check_kfunc_ids)
794 BTF_ID_FLAGS(func, bpf_kfunc_call_test1)
795 BTF_ID_FLAGS(func, bpf_kfunc_call_test2)
796 BTF_ID_FLAGS(func, bpf_kfunc_call_test3)
797 BTF_ID_FLAGS(func, bpf_kfunc_call_test4)
798 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acquire, KF_ACQUIRE | KF_RET_NULL)
799 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_acquire, KF_ACQUIRE | KF_RET_NULL)
800 BTF_ID_FLAGS(func, bpf_kfunc_call_test_release, KF_RELEASE)
801 BTF_ID_FLAGS(func, bpf_kfunc_call_memb_release, KF_RELEASE)
802 BTF_ID_FLAGS(func, bpf_kfunc_call_memb1_release, KF_RELEASE)
803 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdwr_mem, KF_RET_NULL)
804 BTF_ID_FLAGS(func, bpf_kfunc_call_test_get_rdonly_mem, KF_RET_NULL)
805 BTF_ID_FLAGS(func, bpf_kfunc_call_test_acq_rdonly_mem, KF_ACQUIRE | KF_RET_NULL)
806 BTF_ID_FLAGS(func, bpf_kfunc_call_int_mem_release, KF_RELEASE)
807 BTF_ID_FLAGS(func, bpf_kfunc_call_test_kptr_get, KF_ACQUIRE | KF_RET_NULL | KF_KPTR_GET)
808 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass_ctx)
809 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass1)
810 BTF_ID_FLAGS(func, bpf_kfunc_call_test_pass2)
811 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail1)
812 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail2)
813 BTF_ID_FLAGS(func, bpf_kfunc_call_test_fail3)
814 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_pass1)
815 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail1)
816 BTF_ID_FLAGS(func, bpf_kfunc_call_test_mem_len_fail2)
817 BTF_ID_FLAGS(func, bpf_kfunc_call_test_ref, KF_TRUSTED_ARGS | KF_RCU)
818 BTF_ID_FLAGS(func, bpf_kfunc_call_test_destructive, KF_DESTRUCTIVE)
819 BTF_ID_FLAGS(func, bpf_kfunc_call_test_static_unused_arg)
820 BTF_ID_FLAGS(func, bpf_kfunc_call_test_offset)
821 BTF_SET8_END(test_sk_check_kfunc_ids)
822 
823 static void *bpf_test_init(const union bpf_attr *kattr, u32 user_size,
824 			   u32 size, u32 headroom, u32 tailroom)
825 {
826 	void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
827 	void *data;
828 
829 	if (size < ETH_HLEN || size > PAGE_SIZE - headroom - tailroom)
830 		return ERR_PTR(-EINVAL);
831 
832 	if (user_size > size)
833 		return ERR_PTR(-EMSGSIZE);
834 
835 	size = SKB_DATA_ALIGN(size);
836 	data = kzalloc(size + headroom + tailroom, GFP_USER);
837 	if (!data)
838 		return ERR_PTR(-ENOMEM);
839 
840 	if (copy_from_user(data + headroom, data_in, user_size)) {
841 		kfree(data);
842 		return ERR_PTR(-EFAULT);
843 	}
844 
845 	return data;
846 }
847 
848 int bpf_prog_test_run_tracing(struct bpf_prog *prog,
849 			      const union bpf_attr *kattr,
850 			      union bpf_attr __user *uattr)
851 {
852 	struct bpf_fentry_test_t arg = {};
853 	u16 side_effect = 0, ret = 0;
854 	int b = 2, err = -EFAULT;
855 	u32 retval = 0;
856 
857 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
858 		return -EINVAL;
859 
860 	switch (prog->expected_attach_type) {
861 	case BPF_TRACE_FENTRY:
862 	case BPF_TRACE_FEXIT:
863 		if (bpf_fentry_test1(1) != 2 ||
864 		    bpf_fentry_test2(2, 3) != 5 ||
865 		    bpf_fentry_test3(4, 5, 6) != 15 ||
866 		    bpf_fentry_test4((void *)7, 8, 9, 10) != 34 ||
867 		    bpf_fentry_test5(11, (void *)12, 13, 14, 15) != 65 ||
868 		    bpf_fentry_test6(16, (void *)17, 18, 19, (void *)20, 21) != 111 ||
869 		    bpf_fentry_test7((struct bpf_fentry_test_t *)0) != 0 ||
870 		    bpf_fentry_test8(&arg) != 0 ||
871 		    bpf_fentry_test9(&retval) != 0)
872 			goto out;
873 		break;
874 	case BPF_MODIFY_RETURN:
875 		ret = bpf_modify_return_test(1, &b);
876 		if (b != 2)
877 			side_effect = 1;
878 		break;
879 	default:
880 		goto out;
881 	}
882 
883 	retval = ((u32)side_effect << 16) | ret;
884 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(retval)))
885 		goto out;
886 
887 	err = 0;
888 out:
889 	trace_bpf_test_finish(&err);
890 	return err;
891 }
892 
893 struct bpf_raw_tp_test_run_info {
894 	struct bpf_prog *prog;
895 	void *ctx;
896 	u32 retval;
897 };
898 
899 static void
900 __bpf_prog_test_run_raw_tp(void *data)
901 {
902 	struct bpf_raw_tp_test_run_info *info = data;
903 
904 	rcu_read_lock();
905 	info->retval = bpf_prog_run(info->prog, info->ctx);
906 	rcu_read_unlock();
907 }
908 
909 int bpf_prog_test_run_raw_tp(struct bpf_prog *prog,
910 			     const union bpf_attr *kattr,
911 			     union bpf_attr __user *uattr)
912 {
913 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
914 	__u32 ctx_size_in = kattr->test.ctx_size_in;
915 	struct bpf_raw_tp_test_run_info info;
916 	int cpu = kattr->test.cpu, err = 0;
917 	int current_cpu;
918 
919 	/* doesn't support data_in/out, ctx_out, duration, or repeat */
920 	if (kattr->test.data_in || kattr->test.data_out ||
921 	    kattr->test.ctx_out || kattr->test.duration ||
922 	    kattr->test.repeat || kattr->test.batch_size)
923 		return -EINVAL;
924 
925 	if (ctx_size_in < prog->aux->max_ctx_offset ||
926 	    ctx_size_in > MAX_BPF_FUNC_ARGS * sizeof(u64))
927 		return -EINVAL;
928 
929 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 && cpu != 0)
930 		return -EINVAL;
931 
932 	if (ctx_size_in) {
933 		info.ctx = memdup_user(ctx_in, ctx_size_in);
934 		if (IS_ERR(info.ctx))
935 			return PTR_ERR(info.ctx);
936 	} else {
937 		info.ctx = NULL;
938 	}
939 
940 	info.prog = prog;
941 
942 	current_cpu = get_cpu();
943 	if ((kattr->test.flags & BPF_F_TEST_RUN_ON_CPU) == 0 ||
944 	    cpu == current_cpu) {
945 		__bpf_prog_test_run_raw_tp(&info);
946 	} else if (cpu >= nr_cpu_ids || !cpu_online(cpu)) {
947 		/* smp_call_function_single() also checks cpu_online()
948 		 * after csd_lock(). However, since cpu is from user
949 		 * space, let's do an extra quick check to filter out
950 		 * invalid value before smp_call_function_single().
951 		 */
952 		err = -ENXIO;
953 	} else {
954 		err = smp_call_function_single(cpu, __bpf_prog_test_run_raw_tp,
955 					       &info, 1);
956 	}
957 	put_cpu();
958 
959 	if (!err &&
960 	    copy_to_user(&uattr->test.retval, &info.retval, sizeof(u32)))
961 		err = -EFAULT;
962 
963 	kfree(info.ctx);
964 	return err;
965 }
966 
967 static void *bpf_ctx_init(const union bpf_attr *kattr, u32 max_size)
968 {
969 	void __user *data_in = u64_to_user_ptr(kattr->test.ctx_in);
970 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
971 	u32 size = kattr->test.ctx_size_in;
972 	void *data;
973 	int err;
974 
975 	if (!data_in && !data_out)
976 		return NULL;
977 
978 	data = kzalloc(max_size, GFP_USER);
979 	if (!data)
980 		return ERR_PTR(-ENOMEM);
981 
982 	if (data_in) {
983 		err = bpf_check_uarg_tail_zero(USER_BPFPTR(data_in), max_size, size);
984 		if (err) {
985 			kfree(data);
986 			return ERR_PTR(err);
987 		}
988 
989 		size = min_t(u32, max_size, size);
990 		if (copy_from_user(data, data_in, size)) {
991 			kfree(data);
992 			return ERR_PTR(-EFAULT);
993 		}
994 	}
995 	return data;
996 }
997 
998 static int bpf_ctx_finish(const union bpf_attr *kattr,
999 			  union bpf_attr __user *uattr, const void *data,
1000 			  u32 size)
1001 {
1002 	void __user *data_out = u64_to_user_ptr(kattr->test.ctx_out);
1003 	int err = -EFAULT;
1004 	u32 copy_size = size;
1005 
1006 	if (!data || !data_out)
1007 		return 0;
1008 
1009 	if (copy_size > kattr->test.ctx_size_out) {
1010 		copy_size = kattr->test.ctx_size_out;
1011 		err = -ENOSPC;
1012 	}
1013 
1014 	if (copy_to_user(data_out, data, copy_size))
1015 		goto out;
1016 	if (copy_to_user(&uattr->test.ctx_size_out, &size, sizeof(size)))
1017 		goto out;
1018 	if (err != -ENOSPC)
1019 		err = 0;
1020 out:
1021 	return err;
1022 }
1023 
1024 /**
1025  * range_is_zero - test whether buffer is initialized
1026  * @buf: buffer to check
1027  * @from: check from this position
1028  * @to: check up until (excluding) this position
1029  *
1030  * This function returns true if the there is a non-zero byte
1031  * in the buf in the range [from,to).
1032  */
1033 static inline bool range_is_zero(void *buf, size_t from, size_t to)
1034 {
1035 	return !memchr_inv((u8 *)buf + from, 0, to - from);
1036 }
1037 
1038 static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb)
1039 {
1040 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
1041 
1042 	if (!__skb)
1043 		return 0;
1044 
1045 	/* make sure the fields we don't use are zeroed */
1046 	if (!range_is_zero(__skb, 0, offsetof(struct __sk_buff, mark)))
1047 		return -EINVAL;
1048 
1049 	/* mark is allowed */
1050 
1051 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, mark),
1052 			   offsetof(struct __sk_buff, priority)))
1053 		return -EINVAL;
1054 
1055 	/* priority is allowed */
1056 	/* ingress_ifindex is allowed */
1057 	/* ifindex is allowed */
1058 
1059 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, ifindex),
1060 			   offsetof(struct __sk_buff, cb)))
1061 		return -EINVAL;
1062 
1063 	/* cb is allowed */
1064 
1065 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, cb),
1066 			   offsetof(struct __sk_buff, tstamp)))
1067 		return -EINVAL;
1068 
1069 	/* tstamp is allowed */
1070 	/* wire_len is allowed */
1071 	/* gso_segs is allowed */
1072 
1073 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_segs),
1074 			   offsetof(struct __sk_buff, gso_size)))
1075 		return -EINVAL;
1076 
1077 	/* gso_size is allowed */
1078 
1079 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, gso_size),
1080 			   offsetof(struct __sk_buff, hwtstamp)))
1081 		return -EINVAL;
1082 
1083 	/* hwtstamp is allowed */
1084 
1085 	if (!range_is_zero(__skb, offsetofend(struct __sk_buff, hwtstamp),
1086 			   sizeof(struct __sk_buff)))
1087 		return -EINVAL;
1088 
1089 	skb->mark = __skb->mark;
1090 	skb->priority = __skb->priority;
1091 	skb->skb_iif = __skb->ingress_ifindex;
1092 	skb->tstamp = __skb->tstamp;
1093 	memcpy(&cb->data, __skb->cb, QDISC_CB_PRIV_LEN);
1094 
1095 	if (__skb->wire_len == 0) {
1096 		cb->pkt_len = skb->len;
1097 	} else {
1098 		if (__skb->wire_len < skb->len ||
1099 		    __skb->wire_len > GSO_LEGACY_MAX_SIZE)
1100 			return -EINVAL;
1101 		cb->pkt_len = __skb->wire_len;
1102 	}
1103 
1104 	if (__skb->gso_segs > GSO_MAX_SEGS)
1105 		return -EINVAL;
1106 	skb_shinfo(skb)->gso_segs = __skb->gso_segs;
1107 	skb_shinfo(skb)->gso_size = __skb->gso_size;
1108 	skb_shinfo(skb)->hwtstamps.hwtstamp = __skb->hwtstamp;
1109 
1110 	return 0;
1111 }
1112 
1113 static void convert_skb_to___skb(struct sk_buff *skb, struct __sk_buff *__skb)
1114 {
1115 	struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb;
1116 
1117 	if (!__skb)
1118 		return;
1119 
1120 	__skb->mark = skb->mark;
1121 	__skb->priority = skb->priority;
1122 	__skb->ingress_ifindex = skb->skb_iif;
1123 	__skb->ifindex = skb->dev->ifindex;
1124 	__skb->tstamp = skb->tstamp;
1125 	memcpy(__skb->cb, &cb->data, QDISC_CB_PRIV_LEN);
1126 	__skb->wire_len = cb->pkt_len;
1127 	__skb->gso_segs = skb_shinfo(skb)->gso_segs;
1128 	__skb->hwtstamp = skb_shinfo(skb)->hwtstamps.hwtstamp;
1129 }
1130 
1131 static struct proto bpf_dummy_proto = {
1132 	.name   = "bpf_dummy",
1133 	.owner  = THIS_MODULE,
1134 	.obj_size = sizeof(struct sock),
1135 };
1136 
1137 int bpf_prog_test_run_skb(struct bpf_prog *prog, const union bpf_attr *kattr,
1138 			  union bpf_attr __user *uattr)
1139 {
1140 	bool is_l2 = false, is_direct_pkt_access = false;
1141 	struct net *net = current->nsproxy->net_ns;
1142 	struct net_device *dev = net->loopback_dev;
1143 	u32 size = kattr->test.data_size_in;
1144 	u32 repeat = kattr->test.repeat;
1145 	struct __sk_buff *ctx = NULL;
1146 	u32 retval, duration;
1147 	int hh_len = ETH_HLEN;
1148 	struct sk_buff *skb;
1149 	struct sock *sk;
1150 	void *data;
1151 	int ret;
1152 
1153 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1154 		return -EINVAL;
1155 
1156 	data = bpf_test_init(kattr, kattr->test.data_size_in,
1157 			     size, NET_SKB_PAD + NET_IP_ALIGN,
1158 			     SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
1159 	if (IS_ERR(data))
1160 		return PTR_ERR(data);
1161 
1162 	ctx = bpf_ctx_init(kattr, sizeof(struct __sk_buff));
1163 	if (IS_ERR(ctx)) {
1164 		kfree(data);
1165 		return PTR_ERR(ctx);
1166 	}
1167 
1168 	switch (prog->type) {
1169 	case BPF_PROG_TYPE_SCHED_CLS:
1170 	case BPF_PROG_TYPE_SCHED_ACT:
1171 		is_l2 = true;
1172 		fallthrough;
1173 	case BPF_PROG_TYPE_LWT_IN:
1174 	case BPF_PROG_TYPE_LWT_OUT:
1175 	case BPF_PROG_TYPE_LWT_XMIT:
1176 		is_direct_pkt_access = true;
1177 		break;
1178 	default:
1179 		break;
1180 	}
1181 
1182 	sk = sk_alloc(net, AF_UNSPEC, GFP_USER, &bpf_dummy_proto, 1);
1183 	if (!sk) {
1184 		kfree(data);
1185 		kfree(ctx);
1186 		return -ENOMEM;
1187 	}
1188 	sock_init_data(NULL, sk);
1189 
1190 	skb = slab_build_skb(data);
1191 	if (!skb) {
1192 		kfree(data);
1193 		kfree(ctx);
1194 		sk_free(sk);
1195 		return -ENOMEM;
1196 	}
1197 	skb->sk = sk;
1198 
1199 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1200 	__skb_put(skb, size);
1201 	if (ctx && ctx->ifindex > 1) {
1202 		dev = dev_get_by_index(net, ctx->ifindex);
1203 		if (!dev) {
1204 			ret = -ENODEV;
1205 			goto out;
1206 		}
1207 	}
1208 	skb->protocol = eth_type_trans(skb, dev);
1209 	skb_reset_network_header(skb);
1210 
1211 	switch (skb->protocol) {
1212 	case htons(ETH_P_IP):
1213 		sk->sk_family = AF_INET;
1214 		if (sizeof(struct iphdr) <= skb_headlen(skb)) {
1215 			sk->sk_rcv_saddr = ip_hdr(skb)->saddr;
1216 			sk->sk_daddr = ip_hdr(skb)->daddr;
1217 		}
1218 		break;
1219 #if IS_ENABLED(CONFIG_IPV6)
1220 	case htons(ETH_P_IPV6):
1221 		sk->sk_family = AF_INET6;
1222 		if (sizeof(struct ipv6hdr) <= skb_headlen(skb)) {
1223 			sk->sk_v6_rcv_saddr = ipv6_hdr(skb)->saddr;
1224 			sk->sk_v6_daddr = ipv6_hdr(skb)->daddr;
1225 		}
1226 		break;
1227 #endif
1228 	default:
1229 		break;
1230 	}
1231 
1232 	if (is_l2)
1233 		__skb_push(skb, hh_len);
1234 	if (is_direct_pkt_access)
1235 		bpf_compute_data_pointers(skb);
1236 	ret = convert___skb_to_skb(skb, ctx);
1237 	if (ret)
1238 		goto out;
1239 	ret = bpf_test_run(prog, skb, repeat, &retval, &duration, false);
1240 	if (ret)
1241 		goto out;
1242 	if (!is_l2) {
1243 		if (skb_headroom(skb) < hh_len) {
1244 			int nhead = HH_DATA_ALIGN(hh_len - skb_headroom(skb));
1245 
1246 			if (pskb_expand_head(skb, nhead, 0, GFP_USER)) {
1247 				ret = -ENOMEM;
1248 				goto out;
1249 			}
1250 		}
1251 		memset(__skb_push(skb, hh_len), 0, hh_len);
1252 	}
1253 	convert_skb_to___skb(skb, ctx);
1254 
1255 	size = skb->len;
1256 	/* bpf program can never convert linear skb to non-linear */
1257 	if (WARN_ON_ONCE(skb_is_nonlinear(skb)))
1258 		size = skb_headlen(skb);
1259 	ret = bpf_test_finish(kattr, uattr, skb->data, NULL, size, retval,
1260 			      duration);
1261 	if (!ret)
1262 		ret = bpf_ctx_finish(kattr, uattr, ctx,
1263 				     sizeof(struct __sk_buff));
1264 out:
1265 	if (dev && dev != net->loopback_dev)
1266 		dev_put(dev);
1267 	kfree_skb(skb);
1268 	sk_free(sk);
1269 	kfree(ctx);
1270 	return ret;
1271 }
1272 
1273 static int xdp_convert_md_to_buff(struct xdp_md *xdp_md, struct xdp_buff *xdp)
1274 {
1275 	unsigned int ingress_ifindex, rx_queue_index;
1276 	struct netdev_rx_queue *rxqueue;
1277 	struct net_device *device;
1278 
1279 	if (!xdp_md)
1280 		return 0;
1281 
1282 	if (xdp_md->egress_ifindex != 0)
1283 		return -EINVAL;
1284 
1285 	ingress_ifindex = xdp_md->ingress_ifindex;
1286 	rx_queue_index = xdp_md->rx_queue_index;
1287 
1288 	if (!ingress_ifindex && rx_queue_index)
1289 		return -EINVAL;
1290 
1291 	if (ingress_ifindex) {
1292 		device = dev_get_by_index(current->nsproxy->net_ns,
1293 					  ingress_ifindex);
1294 		if (!device)
1295 			return -ENODEV;
1296 
1297 		if (rx_queue_index >= device->real_num_rx_queues)
1298 			goto free_dev;
1299 
1300 		rxqueue = __netif_get_rx_queue(device, rx_queue_index);
1301 
1302 		if (!xdp_rxq_info_is_reg(&rxqueue->xdp_rxq))
1303 			goto free_dev;
1304 
1305 		xdp->rxq = &rxqueue->xdp_rxq;
1306 		/* The device is now tracked in the xdp->rxq for later
1307 		 * dev_put()
1308 		 */
1309 	}
1310 
1311 	xdp->data = xdp->data_meta + xdp_md->data;
1312 	return 0;
1313 
1314 free_dev:
1315 	dev_put(device);
1316 	return -EINVAL;
1317 }
1318 
1319 static void xdp_convert_buff_to_md(struct xdp_buff *xdp, struct xdp_md *xdp_md)
1320 {
1321 	if (!xdp_md)
1322 		return;
1323 
1324 	xdp_md->data = xdp->data - xdp->data_meta;
1325 	xdp_md->data_end = xdp->data_end - xdp->data_meta;
1326 
1327 	if (xdp_md->ingress_ifindex)
1328 		dev_put(xdp->rxq->dev);
1329 }
1330 
1331 int bpf_prog_test_run_xdp(struct bpf_prog *prog, const union bpf_attr *kattr,
1332 			  union bpf_attr __user *uattr)
1333 {
1334 	bool do_live = (kattr->test.flags & BPF_F_TEST_XDP_LIVE_FRAMES);
1335 	u32 tailroom = SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1336 	u32 batch_size = kattr->test.batch_size;
1337 	u32 retval = 0, duration, max_data_sz;
1338 	u32 size = kattr->test.data_size_in;
1339 	u32 headroom = XDP_PACKET_HEADROOM;
1340 	u32 repeat = kattr->test.repeat;
1341 	struct netdev_rx_queue *rxqueue;
1342 	struct skb_shared_info *sinfo;
1343 	struct xdp_buff xdp = {};
1344 	int i, ret = -EINVAL;
1345 	struct xdp_md *ctx;
1346 	void *data;
1347 
1348 	if (prog->expected_attach_type == BPF_XDP_DEVMAP ||
1349 	    prog->expected_attach_type == BPF_XDP_CPUMAP)
1350 		return -EINVAL;
1351 
1352 	if (kattr->test.flags & ~BPF_F_TEST_XDP_LIVE_FRAMES)
1353 		return -EINVAL;
1354 
1355 	if (bpf_prog_is_dev_bound(prog->aux))
1356 		return -EINVAL;
1357 
1358 	if (do_live) {
1359 		if (!batch_size)
1360 			batch_size = NAPI_POLL_WEIGHT;
1361 		else if (batch_size > TEST_XDP_MAX_BATCH)
1362 			return -E2BIG;
1363 
1364 		headroom += sizeof(struct xdp_page_head);
1365 	} else if (batch_size) {
1366 		return -EINVAL;
1367 	}
1368 
1369 	ctx = bpf_ctx_init(kattr, sizeof(struct xdp_md));
1370 	if (IS_ERR(ctx))
1371 		return PTR_ERR(ctx);
1372 
1373 	if (ctx) {
1374 		/* There can't be user provided data before the meta data */
1375 		if (ctx->data_meta || ctx->data_end != size ||
1376 		    ctx->data > ctx->data_end ||
1377 		    unlikely(xdp_metalen_invalid(ctx->data)) ||
1378 		    (do_live && (kattr->test.data_out || kattr->test.ctx_out)))
1379 			goto free_ctx;
1380 		/* Meta data is allocated from the headroom */
1381 		headroom -= ctx->data;
1382 	}
1383 
1384 	max_data_sz = 4096 - headroom - tailroom;
1385 	if (size > max_data_sz) {
1386 		/* disallow live data mode for jumbo frames */
1387 		if (do_live)
1388 			goto free_ctx;
1389 		size = max_data_sz;
1390 	}
1391 
1392 	data = bpf_test_init(kattr, size, max_data_sz, headroom, tailroom);
1393 	if (IS_ERR(data)) {
1394 		ret = PTR_ERR(data);
1395 		goto free_ctx;
1396 	}
1397 
1398 	rxqueue = __netif_get_rx_queue(current->nsproxy->net_ns->loopback_dev, 0);
1399 	rxqueue->xdp_rxq.frag_size = headroom + max_data_sz + tailroom;
1400 	xdp_init_buff(&xdp, rxqueue->xdp_rxq.frag_size, &rxqueue->xdp_rxq);
1401 	xdp_prepare_buff(&xdp, data, headroom, size, true);
1402 	sinfo = xdp_get_shared_info_from_buff(&xdp);
1403 
1404 	ret = xdp_convert_md_to_buff(ctx, &xdp);
1405 	if (ret)
1406 		goto free_data;
1407 
1408 	if (unlikely(kattr->test.data_size_in > size)) {
1409 		void __user *data_in = u64_to_user_ptr(kattr->test.data_in);
1410 
1411 		while (size < kattr->test.data_size_in) {
1412 			struct page *page;
1413 			skb_frag_t *frag;
1414 			u32 data_len;
1415 
1416 			if (sinfo->nr_frags == MAX_SKB_FRAGS) {
1417 				ret = -ENOMEM;
1418 				goto out;
1419 			}
1420 
1421 			page = alloc_page(GFP_KERNEL);
1422 			if (!page) {
1423 				ret = -ENOMEM;
1424 				goto out;
1425 			}
1426 
1427 			frag = &sinfo->frags[sinfo->nr_frags++];
1428 			__skb_frag_set_page(frag, page);
1429 
1430 			data_len = min_t(u32, kattr->test.data_size_in - size,
1431 					 PAGE_SIZE);
1432 			skb_frag_size_set(frag, data_len);
1433 
1434 			if (copy_from_user(page_address(page), data_in + size,
1435 					   data_len)) {
1436 				ret = -EFAULT;
1437 				goto out;
1438 			}
1439 			sinfo->xdp_frags_size += data_len;
1440 			size += data_len;
1441 		}
1442 		xdp_buff_set_frags_flag(&xdp);
1443 	}
1444 
1445 	if (repeat > 1)
1446 		bpf_prog_change_xdp(NULL, prog);
1447 
1448 	if (do_live)
1449 		ret = bpf_test_run_xdp_live(prog, &xdp, repeat, batch_size, &duration);
1450 	else
1451 		ret = bpf_test_run(prog, &xdp, repeat, &retval, &duration, true);
1452 	/* We convert the xdp_buff back to an xdp_md before checking the return
1453 	 * code so the reference count of any held netdevice will be decremented
1454 	 * even if the test run failed.
1455 	 */
1456 	xdp_convert_buff_to_md(&xdp, ctx);
1457 	if (ret)
1458 		goto out;
1459 
1460 	size = xdp.data_end - xdp.data_meta + sinfo->xdp_frags_size;
1461 	ret = bpf_test_finish(kattr, uattr, xdp.data_meta, sinfo, size,
1462 			      retval, duration);
1463 	if (!ret)
1464 		ret = bpf_ctx_finish(kattr, uattr, ctx,
1465 				     sizeof(struct xdp_md));
1466 
1467 out:
1468 	if (repeat > 1)
1469 		bpf_prog_change_xdp(prog, NULL);
1470 free_data:
1471 	for (i = 0; i < sinfo->nr_frags; i++)
1472 		__free_page(skb_frag_page(&sinfo->frags[i]));
1473 	kfree(data);
1474 free_ctx:
1475 	kfree(ctx);
1476 	return ret;
1477 }
1478 
1479 static int verify_user_bpf_flow_keys(struct bpf_flow_keys *ctx)
1480 {
1481 	/* make sure the fields we don't use are zeroed */
1482 	if (!range_is_zero(ctx, 0, offsetof(struct bpf_flow_keys, flags)))
1483 		return -EINVAL;
1484 
1485 	/* flags is allowed */
1486 
1487 	if (!range_is_zero(ctx, offsetofend(struct bpf_flow_keys, flags),
1488 			   sizeof(struct bpf_flow_keys)))
1489 		return -EINVAL;
1490 
1491 	return 0;
1492 }
1493 
1494 int bpf_prog_test_run_flow_dissector(struct bpf_prog *prog,
1495 				     const union bpf_attr *kattr,
1496 				     union bpf_attr __user *uattr)
1497 {
1498 	struct bpf_test_timer t = { NO_PREEMPT };
1499 	u32 size = kattr->test.data_size_in;
1500 	struct bpf_flow_dissector ctx = {};
1501 	u32 repeat = kattr->test.repeat;
1502 	struct bpf_flow_keys *user_ctx;
1503 	struct bpf_flow_keys flow_keys;
1504 	const struct ethhdr *eth;
1505 	unsigned int flags = 0;
1506 	u32 retval, duration;
1507 	void *data;
1508 	int ret;
1509 
1510 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1511 		return -EINVAL;
1512 
1513 	if (size < ETH_HLEN)
1514 		return -EINVAL;
1515 
1516 	data = bpf_test_init(kattr, kattr->test.data_size_in, size, 0, 0);
1517 	if (IS_ERR(data))
1518 		return PTR_ERR(data);
1519 
1520 	eth = (struct ethhdr *)data;
1521 
1522 	if (!repeat)
1523 		repeat = 1;
1524 
1525 	user_ctx = bpf_ctx_init(kattr, sizeof(struct bpf_flow_keys));
1526 	if (IS_ERR(user_ctx)) {
1527 		kfree(data);
1528 		return PTR_ERR(user_ctx);
1529 	}
1530 	if (user_ctx) {
1531 		ret = verify_user_bpf_flow_keys(user_ctx);
1532 		if (ret)
1533 			goto out;
1534 		flags = user_ctx->flags;
1535 	}
1536 
1537 	ctx.flow_keys = &flow_keys;
1538 	ctx.data = data;
1539 	ctx.data_end = (__u8 *)data + size;
1540 
1541 	bpf_test_timer_enter(&t);
1542 	do {
1543 		retval = bpf_flow_dissect(prog, &ctx, eth->h_proto, ETH_HLEN,
1544 					  size, flags);
1545 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1546 	bpf_test_timer_leave(&t);
1547 
1548 	if (ret < 0)
1549 		goto out;
1550 
1551 	ret = bpf_test_finish(kattr, uattr, &flow_keys, NULL,
1552 			      sizeof(flow_keys), retval, duration);
1553 	if (!ret)
1554 		ret = bpf_ctx_finish(kattr, uattr, user_ctx,
1555 				     sizeof(struct bpf_flow_keys));
1556 
1557 out:
1558 	kfree(user_ctx);
1559 	kfree(data);
1560 	return ret;
1561 }
1562 
1563 int bpf_prog_test_run_sk_lookup(struct bpf_prog *prog, const union bpf_attr *kattr,
1564 				union bpf_attr __user *uattr)
1565 {
1566 	struct bpf_test_timer t = { NO_PREEMPT };
1567 	struct bpf_prog_array *progs = NULL;
1568 	struct bpf_sk_lookup_kern ctx = {};
1569 	u32 repeat = kattr->test.repeat;
1570 	struct bpf_sk_lookup *user_ctx;
1571 	u32 retval, duration;
1572 	int ret = -EINVAL;
1573 
1574 	if (kattr->test.flags || kattr->test.cpu || kattr->test.batch_size)
1575 		return -EINVAL;
1576 
1577 	if (kattr->test.data_in || kattr->test.data_size_in || kattr->test.data_out ||
1578 	    kattr->test.data_size_out)
1579 		return -EINVAL;
1580 
1581 	if (!repeat)
1582 		repeat = 1;
1583 
1584 	user_ctx = bpf_ctx_init(kattr, sizeof(*user_ctx));
1585 	if (IS_ERR(user_ctx))
1586 		return PTR_ERR(user_ctx);
1587 
1588 	if (!user_ctx)
1589 		return -EINVAL;
1590 
1591 	if (user_ctx->sk)
1592 		goto out;
1593 
1594 	if (!range_is_zero(user_ctx, offsetofend(typeof(*user_ctx), local_port), sizeof(*user_ctx)))
1595 		goto out;
1596 
1597 	if (user_ctx->local_port > U16_MAX) {
1598 		ret = -ERANGE;
1599 		goto out;
1600 	}
1601 
1602 	ctx.family = (u16)user_ctx->family;
1603 	ctx.protocol = (u16)user_ctx->protocol;
1604 	ctx.dport = (u16)user_ctx->local_port;
1605 	ctx.sport = user_ctx->remote_port;
1606 
1607 	switch (ctx.family) {
1608 	case AF_INET:
1609 		ctx.v4.daddr = (__force __be32)user_ctx->local_ip4;
1610 		ctx.v4.saddr = (__force __be32)user_ctx->remote_ip4;
1611 		break;
1612 
1613 #if IS_ENABLED(CONFIG_IPV6)
1614 	case AF_INET6:
1615 		ctx.v6.daddr = (struct in6_addr *)user_ctx->local_ip6;
1616 		ctx.v6.saddr = (struct in6_addr *)user_ctx->remote_ip6;
1617 		break;
1618 #endif
1619 
1620 	default:
1621 		ret = -EAFNOSUPPORT;
1622 		goto out;
1623 	}
1624 
1625 	progs = bpf_prog_array_alloc(1, GFP_KERNEL);
1626 	if (!progs) {
1627 		ret = -ENOMEM;
1628 		goto out;
1629 	}
1630 
1631 	progs->items[0].prog = prog;
1632 
1633 	bpf_test_timer_enter(&t);
1634 	do {
1635 		ctx.selected_sk = NULL;
1636 		retval = BPF_PROG_SK_LOOKUP_RUN_ARRAY(progs, ctx, bpf_prog_run);
1637 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, &duration));
1638 	bpf_test_timer_leave(&t);
1639 
1640 	if (ret < 0)
1641 		goto out;
1642 
1643 	user_ctx->cookie = 0;
1644 	if (ctx.selected_sk) {
1645 		if (ctx.selected_sk->sk_reuseport && !ctx.no_reuseport) {
1646 			ret = -EOPNOTSUPP;
1647 			goto out;
1648 		}
1649 
1650 		user_ctx->cookie = sock_gen_cookie(ctx.selected_sk);
1651 	}
1652 
1653 	ret = bpf_test_finish(kattr, uattr, NULL, NULL, 0, retval, duration);
1654 	if (!ret)
1655 		ret = bpf_ctx_finish(kattr, uattr, user_ctx, sizeof(*user_ctx));
1656 
1657 out:
1658 	bpf_prog_array_free(progs);
1659 	kfree(user_ctx);
1660 	return ret;
1661 }
1662 
1663 int bpf_prog_test_run_syscall(struct bpf_prog *prog,
1664 			      const union bpf_attr *kattr,
1665 			      union bpf_attr __user *uattr)
1666 {
1667 	void __user *ctx_in = u64_to_user_ptr(kattr->test.ctx_in);
1668 	__u32 ctx_size_in = kattr->test.ctx_size_in;
1669 	void *ctx = NULL;
1670 	u32 retval;
1671 	int err = 0;
1672 
1673 	/* doesn't support data_in/out, ctx_out, duration, or repeat or flags */
1674 	if (kattr->test.data_in || kattr->test.data_out ||
1675 	    kattr->test.ctx_out || kattr->test.duration ||
1676 	    kattr->test.repeat || kattr->test.flags ||
1677 	    kattr->test.batch_size)
1678 		return -EINVAL;
1679 
1680 	if (ctx_size_in < prog->aux->max_ctx_offset ||
1681 	    ctx_size_in > U16_MAX)
1682 		return -EINVAL;
1683 
1684 	if (ctx_size_in) {
1685 		ctx = memdup_user(ctx_in, ctx_size_in);
1686 		if (IS_ERR(ctx))
1687 			return PTR_ERR(ctx);
1688 	}
1689 
1690 	rcu_read_lock_trace();
1691 	retval = bpf_prog_run_pin_on_cpu(prog, ctx);
1692 	rcu_read_unlock_trace();
1693 
1694 	if (copy_to_user(&uattr->test.retval, &retval, sizeof(u32))) {
1695 		err = -EFAULT;
1696 		goto out;
1697 	}
1698 	if (ctx_size_in)
1699 		if (copy_to_user(ctx_in, ctx, ctx_size_in))
1700 			err = -EFAULT;
1701 out:
1702 	kfree(ctx);
1703 	return err;
1704 }
1705 
1706 static const struct btf_kfunc_id_set bpf_prog_test_kfunc_set = {
1707 	.owner = THIS_MODULE,
1708 	.set   = &test_sk_check_kfunc_ids,
1709 };
1710 
1711 BTF_ID_LIST(bpf_prog_test_dtor_kfunc_ids)
1712 BTF_ID(struct, prog_test_ref_kfunc)
1713 BTF_ID(func, bpf_kfunc_call_test_release)
1714 BTF_ID(struct, prog_test_member)
1715 BTF_ID(func, bpf_kfunc_call_memb_release)
1716 
1717 static int __init bpf_prog_test_run_init(void)
1718 {
1719 	const struct btf_id_dtor_kfunc bpf_prog_test_dtor_kfunc[] = {
1720 		{
1721 		  .btf_id       = bpf_prog_test_dtor_kfunc_ids[0],
1722 		  .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[1]
1723 		},
1724 		{
1725 		  .btf_id	= bpf_prog_test_dtor_kfunc_ids[2],
1726 		  .kfunc_btf_id = bpf_prog_test_dtor_kfunc_ids[3],
1727 		},
1728 	};
1729 	int ret;
1730 
1731 	ret = register_btf_fmodret_id_set(&bpf_test_modify_return_set);
1732 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_prog_test_kfunc_set);
1733 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, &bpf_prog_test_kfunc_set);
1734 	ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SYSCALL, &bpf_prog_test_kfunc_set);
1735 	return ret ?: register_btf_id_dtor_kfuncs(bpf_prog_test_dtor_kfunc,
1736 						  ARRAY_SIZE(bpf_prog_test_dtor_kfunc),
1737 						  THIS_MODULE);
1738 }
1739 late_initcall(bpf_prog_test_run_init);
1740